LangChain / LlamaIndex retriever integration

[adaamko]

Summary

Wrap VerbatimRAG as a native LangChain BaseRetriever and/or LlamaIndex BaseRetriever so users already on those frameworks can drop verbatim extraction into existing pipelines.

Motivation

Many teams build on LangChain or LlamaIndex. A thin integration layer lets them replace the generation step with verbatim extraction without rewriting their retrieval setup.

Scope

A minimal LangChain integration could look like:

from verbatim_rag.integrations.langchain import VerbatimRetriever

retriever = VerbatimRetriever(rag=rag_system)
docs = retriever.get_relevant_documents("What is the main contribution?")
# Each Document.page_content is a verbatim span; metadata includes source doc

• verbatim_rag/integrations/langchain.py — VerbatimRetriever(BaseRetriever)
• verbatim_rag/integrations/llamaindex.py — VerbatimRetriever(BaseRetriever) (optional, same idea)
• Neither framework needs to be a hard dependency — guard imports with a clear error message
• Short usage example in examples/

Notes

• The integration is a thin adapter; no changes to core logic needed
• langchain-core is the relevant package for BaseRetriever (not full langchain)